Deep Learning and Transformer Models for Groundwater Level Prediction in the Marvdasht Plain: Protecting UNESCO Heritage Sites—Persepolis and Naqsh-e Rustam

A transformer-based deep learning model achieves high accuracy (R² = 0.83, RMSE = 6.15 m) in predicting groundwater levels and subsidence rates near UNESCO World Heritage Sites Persepolis and Naqsh-e Rustam, outperforming traditional deep learning methods by over 40%. The model integrates synthetic aperture radar data with hydrometeorological and biophysical parameters to quantify groundwater stress, revealing critical subsidence risks (exceeding 200 mm/year) that threaten Achaemenid stone reliefs. This approach provides a scalable framework for early-warning systems in heritage conservation, offering actionable insights for sustainable aquifer management.

Heidarian, P., Antezana Lopez, F. P., Tan, Y., Fathtabar Firozjaee, S., Yousefi, T., Salehi, H., Osman Pour, A., Elena Oscori Marca, M., Zhou, G., Azhdari, A., & Shahbazi, R. (2025). Deep Learning and Transformer Models for Groundwater Level Prediction in the Marvdasht Plain: Protecting UNESCO Heritage Sites—Persepolis and Naqsh-e Rustam. Remote Sensing, 17(14), 2532. https://doi.org/10.3390/rs17142532